Breaking

Google Adds Built-In Computer Use to Gemini 3.5 Flash for Autonomous App Control

June 28, 2026 at 23:15 EDT

AI Agents
Foundation Models

On June 24, 2026, Google announced that it had integrated "Computer Use" as a built-in tool in Gemini 3.5 Flash, letting the model view a screen and autonomously click, type and scroll. Developers can now build agents that operate user interfaces from screenshots inside a single model.

June 24, 2026 · Google Gemini

Gemini 3.5 Flash Learns to Operate Your Screen

Google bakes "Computer Use" natively into its low-cost, fast Flash model — letting agents see screenshots, reason, and autonomously click, type and scroll across browser, mobile and desktop.

78.4%

OSWorld-Verified score for computer operation

Turns of capture→reason→act in the demo, fully autonomous

Environments: browser, mobile, desktop (preview)

The agent loop, repeated 73 times

From a screenshot to a finished report — with no human in the seat.

CAPTURE

Take a screenshot of the screen

→

REASON

Recognize the UI and decide next step

→

ACT

Click, type or scroll, then loop back

Big context window, small price

Pricing per 1M tokens — output is 6× the cost of input.

$1.50

Input / 1M

$9.00

Output / 1M

1,048,576 input tokens of context

65,536 output tokens

Why developers cheer

Top-tier UI control at Flash pricing
One model spans browser, mobile and desktop
Combines with Search grounding & Function calling

Handle with care

Test at small scale before production
Keep confirmation prompts enabled
Verify long-term reliability & error recovery

Built-in safeguards

User confirmation is required before irreversible actions, and the model includes detection for prompt injection — malicious input designed to make it misbehave. Available now in preview via Antigravity, the Gemini API and Vertex AI.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.